Understanding the “Understanding of AI”:

Abstract

Recent debates concerning large language models (LLMs) are juxtaposed between claims which hold that such systems either understand language in true semantic ways or that they only do functional simulation. This paper argues that both positions rely on a philosophically inadequate conception of what understanding is or that the age of AI demands a reformulation of the notion of understanding itself. Drawing on functionalist and interpretivist traditions, the paper distinguishes between simulation and semantics proper. LLMs possess minimal understanding in virtue of their functional organization and interpretability which may be read as syntactical and not semantic due to the absence of world-involving triangulation and intrinsic intentionality. All these suggest that consciousness is a sine qua non for understanding which is absent in the understanding of AI systems. It eventually leads to a more nuanced sense of understanding that has much repercussions in the ethico-political concerns arising in the recent discussions of philosophy of mind. Such concerns centre around the issues of ethical subjectivity, responsibility, agency and moral justifications which have political overtones.